Who Cooks the Books in China, and Does It Pay?

Strategic Management Journal Strat. Mgmt. J., 37: 2658–2676 (2016) Published online EarlyView 12 January 2016 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/smj.2466 Received 4 February 2014; Final revision received 10 May 2015

WHO COOKS THE BOOKS IN CHINA, AND DOES IT PAY? EVIDENCE FROM PRIVATE, HIGH-TECHNOLOGY FIRMS TOBY STUART1 and YANBO WANG2* 1 Hass School of Business, UC, Berkeley, Berkeley, California, U.S.A. 2 Cheung Kong Graduate School of Business (CKGSB), Beijing, China

Research summary: We document the extent of fraudulent reporting among 467 private Chinese technology companies. Comparing the financial statements concurrently submitted to two different state agencies, we demonstrate a systematic gap in reported profit figures in the two sets of books. We find: (1) more than half the sampled companies report incentive-compatible, materially discrepant profit numbers to the two agencies; (2) politically connected companies are approximately 18 percent more likely to commit fraud and those with venture capital backing are 19 percent more likely to do so; and (3) it pays to cheat. We estimate that companies who “cook” their books have considerably higher odds of receiving an innovation grant. Especially given its prevalence, we conclude that fraud can be a source of performance differential for emerging market companies. Managerial summary: We document that more than half of a sample of 467 private, Chinese technology companies engage in fraudulent financial reporting. By comparing the financial statements companies concurrently submitted to two different state agencies, we demonstrate a systematic gap in reported profit figures in the two sets of books. Relative to the companies without these attributes, we find that politically connected companies are approximately 18 percent more likely to commit fraud and those with venture capital backing are 19 percent more likely to do so. Furthermore, we show that it pays to cheat. We estimate that companies who “cook” their books have considerably higher odds of receiving a government-sponsored innovation grant. Therefore, fraud can be a source of performance differential for emerging market companies. Copyright © 2015 John Wiley & Sons, Ltd.

INTRODUCTION Schuler, Rehbein and Cramer, 2002; Lester et al., 2008), but it is especially pronounced in emerging A burgeoning literature finds that connections to economies (Li and Zhang, 2007; Malesky and government officials convey advantages to com- Taussig, 2009; Peng and Luo, 2000; Siegel, 2007). panies. The link between political connections and One regrettable channel through which political corporate performance exists in industrialized coun- connections may contribute to corporate perfor- tries (e.g., Hillman, Zardkoohi, and Bierman, 1999; mance is to facilitate profit-enhancing fraud. This occurs because high-level ties to the state may deflect regulatory scrutiny from dubious forms Keywords: financial fraud; political connection; venture of corporate conduct. Even in business environ- capital; emerging market economies ments characterized by strong market-supporting *Correspondence to: Yanbo Wang. Cheung Kong Graduate institutions such as the United States, evidence School of Business, 1 East Chang An Avenue, Beijing, China 100006. E-mail: [email protected] suggests that law enforcement is not universally JEL Classifications: H25, K33, K42, M48 and P16. applied in instances of suspected fraud, which

Copyright © 2015 John Wiley & Sons, Ltd. Who Cooks the Books in China, and Does It Pay? 2659 creates asymmetric payoffs for firms to engage of employees to the next (Baron, Burton, and Han- in fraudulent conduct (Correia, 2009; Fulmer and nan, 1999; Harrison and Carroll, 2006). Identifying Knill, 2012; Yu and Yu, 2011). the precursors of fraud in early-stage companies is In this article, we ask a set of related ques- likely to illuminate the set of companies in which tions: How prevalent is fraud among private, unethical behaviors become cultural inheritances. technology-based companies in China? Do polit- This article uses a unique approach to iden- ical connections and other entrepreneur- and tify cases of fraud among entrepreneurial firms in firm-level characteristics correlate with fraudulent China. Our analysis does not rely on the actions conduct in this sector of the Chinese economy? of regulatory agencies; rather, we directly observe Under what conditions is fraud rewarded? instances of financial data manipulation by collect- We emphasize three findings. First, we quantify ing and comparing two sets of financial books that the effect of political connections on the extent are required to be identical under Chinese law. We of fraud and the ability of firms to access state consider incentive-compatible, discrepant reporting resources. We present a clear illustration of how across the two sets of books to be evidence of finan- political ties translate to a firm-level financial cial fraud. advantage because they enable firms to gain priv- One set of books comes from corporate appli- ileged access to scarce, state-allocated resources. cations for a state-funded innovation grant. To Second, we demonstrate a link between com- apply for this grant applicants must submit pany ownership structure and the propensity to audited financial statements to China’s Ministry commit fraud. Specifically, we (unexpectedly) of Science and Technology (MOST). The MOST find that venture capital-backed companies are decision-makers exhibit a publicly stated preference more likely to perpetrate fraud and to benefit for strong-performing companies, which creates from government-dispensed resources. Finally, we an incentive for applicants to exaggerate profit demonstrate the base rate of fraud in an as-yet levels and to over-state technical achievements. We unexamined segment of the Chinese economy: study companies that applied for the MOST grants the vast sector of young, private, entrepreneurial because, for the same set of companies at the same companies. points in time, we were able to retrieve a second set On the latter point, a sizable literature examines of financial statements. These statements, which cases of large, public company fraud. In conse- were required of all companies in China, were quence of their economic scale, the actions of these submitted to the local State Administration of organizations affect many thousands of employees, Industry and Commerce (SAIC). The SAIC is one investors, suppliers, and customers. Therefore, large of China’s primary regulatory bodies with broad companies are subject to a more focused regula- jurisdiction. Although not directly responsible for tory lens. But because media and regulatory atten- detecting tax evasion, the SAIC does coordinate tion concentrates on large companies, we know with taxation agencies in investigating so-called much less about the incidence of fraud at small, “irregular behaviors” (Tian, 2012). private companies. We believe that understanding Chinese law clearly states that companies must this population is vital. First, in emerging mar- submit the same financial information in these two kets, private firms are primary catalysts of economic sets of books, but many firms may benefit from growth (McMillan and Woodruff, 2002).1 Second, misrepresenting the numbers. In the typical case in major decisions in young companies set in motion our data, companies appear to fall within a range of path-dependent dynamics in which organizational financial performance in which they have incentive practices and values are transmitted from one group to overstate their profitability to the MOST, and possibly to understate it to the SAIC. To be up front about one limitation of the data: We never know the 1For instance, in China between 1998 and 2010, industrial output true numbers for the firms in our sample; we simply by large-firm-dominated state-owned enterprises increased from 2.22 to 6.67 trillion RMB. This compares to an astonishing know whether these companies illegally report dis- increase from 0.32 to 25.23 trillion RMB for the private-company crepant results across their two sets of books. As we sector, which is dominated by small- and medium-sized firms show, however, in all but a handful of cases, report- (National Bureau of Statistics, 1998, 2010). The exchange rate between the Chinese RMB and the U.S. dollar was 0.121 (1 U.S. ing discrepancies are incentive-compatible: Finan- dollar = 8.27 RMB yuan) in 1998, and appreciated to 0.148 (1 U.S. cial performance is almost always stronger in the dollar = 6.77 RMB yuan) in 2010. MOST books, relative to that disclosed to the SAIC.

In addition to providing evidence of the perva- accused of academic plagiarism, and it was only siveness of fraud in China in an as-yet under-studied in the course of investigations of those allegations group of companies, this article offers a final contri- that the firm’s financial misrepresentations were bution. Examinations of fraud typically rely on its discovered (Peng, Li, and Li, 2011). detection for evidence of its existence (e.g., Uzun, This anecdote illustrates three points. First, Szewczyk, and Varma, 2004). In general, fraud is manipulating financial data can lead to access to only observed when perpetrators are caught and the state-controlled resources. Second, fraud in China’s act(s) of fraud are publicized. Researchers typically private sector, although prevalent in our data, does observe detected instances of fraud, but not the carry some risk. We show that cooking the books underlying population of fraudulent acts (Fisman pays, but sanctions do occur. Third, well-connected and Wei’s, 2004, study of tax evasion is a notable actors may be able to delay punishment (or avoid it exception). However, we know from the literature altogether) if their fraudulent conduct is detected. that fraud detection is nonrandom (Greve, Palmer, For this reason, actors whose connections or and Pozner, 2010), and also that entrepreneur and social positions deflect scrutiny may be the most firm characteristics such as political connectedness emboldened to perpetrate fraud. may both negatively affect the detection of fraud China is known to have weak market-supporting and promote its perpetration. If firms with certain institutions. While China’s economy has rapidly attributes are less likely to be investigated by grown during the past few decades, a well- regulatory authorities, there is an obvious bias documented accompaniment to this growth has in estimates of the effect of these attributes on been prevalent corruption (Wademan, 2012). For the incidence of fraud in datasets that depend example, Transparency International, a nongovern- on the detection of fraud for its identification. mental organization that monitors and publicizes The data we have collected enable us to avoid corruption across the globe, consistently has ranked such bias. China high on its corruption index. Although it is a challenge to comprehensively document occurrences of fraud, the literature never- BACKGROUND, THEORY theless is replete with evidence of its existence. For AND HYPOTHESES example, Cumming, Hou, and Lee (2011) identified 604 fraud cases between 1999 and 2008 that were A case illustrates the phenomena we study. The disclosed by the Chinese Stock Exchanges and the politically connected dean of a prominent engineer- China Securities Regulatory Commission (CSRC), ing school founded a company, Taide Compressor, which corresponds to a roughly five percent inci- to commercialize a novel technology based on dence rate of detected cheating in that period.2 university research. Between 2001 and 2005, Taide Among the fraud convictions by the CSRC, a sig- reported a net loss in each year in financial state- nificant proportion involved misleading financial ments submitted to a provincial agency, the SAIC, disclosures (Chen et al., 2006). that is related to the local tax authorities. However, While financial fraud in China is assumed tobe in its application for a state-awarded innovation widespread, law enforcement against it has been grant in 2004, Taide submitted an income statement anemic. Between 2001 and 2006, only 92 listed declaring a net profit of 1.25 million, 2.55 million, companies received CSRC sanctions (Liebman and 14.70 million yuan for 2001, 2002, and and Milhaupt, 2008). Even though it is difficult to 2003, respectively. In reality, Taide’s performance interpret the thoroughness of enforcement activities appeared to have been poor: Despite the claim of in the absence of knowledge of the true baseline profits, illiquidity forced the company to restruc- rate of fraud, the general consensus among scholars ture. The company nevertheless was awarded is that the incidence of official sanctions is modest an innovation grant (Liu, 2011). Ultimately, the relative to the prevalence of false accounting, company’s fraudulent financial claims were dis- insider trading, and inaccurate financial disclosure covered and its award was revoked, but probably because of the founder’s political connections, 2Li et al. (2014) examined the performance of 1,217 firms listed the financial fraud was detected only incidentally on the two main Chinese stock exchanges from 2003 to 2009, and its punishment was delayed by two years. The and found that 33.6 percent of them had assets diverted by their company’s now-discredited founder also stood controlling shareholders in 2005 for “nonoperational” purposes.

Copyright © 2015 John Wiley & Sons, Ltd. Strat. Mgmt. J., 37: 2658–2676 (2016) DOI: 10.1002/smj Who Cooks the Books in China, and Does It Pay? 2661 in China’s equity markets (Liebman and Milhaupt, Consistent with this view, studies have found that 2008; Pistor and Xu, 2005). firms with political ties in developing economies The literature suggests a set of firm-level fac- often secure protections that are difficult to acquire tors influence the incentive to cheat, the riskof through formally sanctioned channels. Faccio fraud detection, and conditional on its discovery, (2006) finds that politically connected firms some- the speed and severity of any sanctions that are times avoid regulatory scrutiny. To the extent politi- meted out. We focus on two, particular firm-level cians provide protection to companies they favor, characteristics: political connectedness and capital connected firms may be more likely to delay or dis- structure. tort information presented in financial disclosures to the firm’s benefit. Likewise, Chaney, Faccio, and Parsley (2011) find that the quality of earnings Political connections reported by connected firms is significantly poorer than that of comparable, nonconnected firms. High-level connections to government officials In China, the influence of political connections convey advantages to firms (e.g., Fisman, 2001; may be amplified by the low base rate of legal Hillman et al., 1999; Li and Zhang, 2007; Peng and enforcement, which enables discretion on the part Luo, 2000; Siegel, 2007). One dimension along of officials concerning the focus of their limited which political connections may contribute to cor- resources for regulatory scrutiny. The institutional, porate performance is by abetting profit-enhancing resource, and political constraints under which the fraud. In the United States, for example, Yu and Yu CSRC operates are manifest in its meager pros- (2011) find that lobbying expenditures significantly ecutorial activities. Given the minimal extent to reduce the risk of fraud detection: Compared to which the CSRC prosecutes fraud, it comes as lit- nonlobbying firms, regulators are much less likely tle surprise that it is hesitant to initiate enforcement to identify acts of fraud perpetrated by lobbying actions against companies with strong ties to the firms. Moreover, if fraud is uncovered, lobbying state. And when actions against them are initiated, firms evade detection for longer periods of time Firth, Rui, and Wu (2012) find that State-Owned and receive lighter sanctions when caught (Fulmer Enterprises (SOEs) have an advantage as defendants and Knill, 2012). Likewise, Correia (2009) finds in court trials against other parties. that politically connected firms are less likely to Turning specifically to smaller firms, given restate their earnings due to a comment letter from that government officials exercise discretion in the Security Exchange Commission (SEC) and are rule enforcement, it is unsurprising that Chinese less likely to become targets of SEC enforcement entrepreneurs perceive regulatory affairs to be one actions. It appears that even in advanced economies of the most significant and least predictable envi- with strong, market-supporting institutions, polit- ronmental factors shaping their business prospects ical engagement may forestall fraud detection and (Tan and Litschert, 1994). Likewise, entrepreneurs reduce the severity of its punishment, conditional consider a connection to the state to be one of the on detection. most important assets for their companies, and There is reason to suspect that the link between they invest significant resources to develop state political connections and the detection and punish- ties. For example, Ma and Parish (2006) show that ment of fraud is stronger in developing economies Chinese entrepreneurs generously donate to gov- characterized by still-nascent legal codes that are ernment welfare projects to gain political access enforced sporadically. Even after three decades of and social status via appointments to political reform, legal enforcement in China still is not trans- councils. Similarly, Nee and Opper (2010) find that parent, which creates latitude for agents to pur- connected firms invest more in “informal contri- sue informal forms of influence with authorities as butions” or outright bribes to state officials. Li and a means to eschew legal scrutiny (e.g., Tan and colleagues further show that such connections pay Litschert, 1994; Xin and Pierce, 1996). Whereas off in terms of company performance, particularly in most developed countries, companies engage in for firms operating in regulated sectors (Li and legalized lobbying activities, in many developing Atuahene-Gima, 2001; Li and Zhang, 2007). countries, Western-style lobbying is banned. In the While many entrepreneurs attempt to influ- absence of legal channels of political influence, ence regulatory oversight, politically connected bribery is commonplace. firms possess an advantage in that they are better

Copyright © 2015 John Wiley & Sons, Ltd. Strat. Mgmt. J., 37: 2658–2676 (2016) DOI: 10.1002/smj 2662 T. Stuart and Y. Wang positioned to engage in gray, and even illegal, reflect poorly on investors and may draw media or practices—including bribery—to avoid scrutiny. regulatory scrutiny. This partially explains why an Because corruption of this nature is punished investment from an external organization often is when it is discovered, it is risky to both partic- accompanied by the formalization of corporate gov- ipants in a bribe. On the respective sides of a ernance practices, such as the creation of boards of would-be transaction, it can be difficult to gauge directors and professionalization of the accounting willingness-to-pay, willingness-to-accept a bribe, function (Baron et al., 1999). ability-to-pay, and capacity-to-deliver promised When the organizational investor is a venture services. Politically connected actors are likely to capital firm (VC), it may be particularly active in have the knowledge and the connections regarding promoting financial transparency at the portfolio whom to bribe, how much to offer for a specific company. To steward their investments and to service, and how to deliver a bribe to minimize risk mitigate agency concerns, VCs insist on multiple to the parties in a transaction. control rights and typically implement extensive In summary, politically connected firms are monitoring and advisory systems as conditions for well positioned to manipulate their accounting making an investment. Through board participation books relative to their unconnected counterparts and a hands-on relationship with senior manage- for several reasons. First, the incentive for a firm ment, VCs often are intimately involved in new to engage in any form of fraudulent behavior ventures, playing roles such as hiring and firing depends on the risk of detection. We anticipate that the CEO (Kaplan and Stromberg, 2003), setting connected firms are less likely to be scrutinized by executive compensation, and electing members regulators. Second, the incentive to perpetrate fraud to the board (Hellmann and Puri, 2002). These depends on the anticipated sanctions, conditional pre-IPO roles dovetail an empirical literature that on detection. Once again, we anticipate that even compares the performance of otherwise similar, if their fraudulent actions are detected, principals post-IPO, VC- and non-VC-backed companies. at politically connected firms are likely to benefit This work typically finds evidence of better gov- from delayed, and possibly reduced, punishments. ernance practices in VC-funded companies (e.g., These considerations lead us to hypothesize: Hochberg, 2012; Morsfield and Tan, 2006). Although work on the governance structures Hypothesis 1: Early-stage technology companies of venture-backed companies admittedly is with one or more politically connected founders U.S.-centric, and there are reasons to believe that are more likely to manipulate their financial VCs in emerging markets are less able to institute data. good governance practices (e.g., Cumming and Walz, 2010; Johan and Zhang, 2014; White, Gao, and Zhang, 2005), we nonetheless hypothesize Ownership and capital structure that VC backing will promote legal compliance. Young companies are heterogeneous in their owner- At minimum, VC firms are repeat players in the ship structures. In particular, some companies have entrepreneurial process. If fraudulent conduct only individual owners, whereas others have orga- on the part of portfolio companies reflects on nizational equity holders, such as venture capital their financial backers, reputational concerns will investors, universities, and other corporations. For encourage VC firms to promote integrity in the several reasons, firms with organizational equity reporting of financial results. These arguments lead holders may be less likely to manipulate their finan- to two, additional hypotheses: cial books. First, firms with organizational investors typically have access to greater resources, and thus, Hypothesis 2: Compared to early-stage technol- face a weaker incentive to commit acts of fraud ogy companies with only individual investors, to acquire additional capital. Second, as outside those with organizational equity owners will be investors, organizational equity holders may be less likely to manipulate their financial data. concerned about the accounting practices in their portfolios because companies that manipulate their Hypothesis 3: Compared to early-stage technol- financial reports to defraud the state also are more ogy companies with organizational equity own- likely to misrepresent their books to investors. In ers, those with venture capital investors will be addition, illegal activities by a portfolio company less likely to manipulate their financial data.

Finally, we do not offer a specific hypothesis, but entrepreneurs’ employment histories and member- in the empirical analysis that follows, we investi- ships in politically influential organizations, which gate whether fraud increases the odds that a firm allows us to construct measures of work experience is awarded state resources. Our arguments for the and political connection. hypotheses rest on the premise that companies have Financial statements filed to the local State an incentive to misreport their financial information Administration of Industry and Commerce (SAIC) to one of two state agencies, often to bolster their were used to compare with the financial statements results in the statements they submit in applications from the MOST. The SAIC is the primary state for a government grant. Therefore, in addition to agency responsible for regulating day-to-day com- being of interest in its own right, the logic under- mercial activities. According to China’s Company lying the hypotheses is strongest if the reporting of Law all commercial entities in the country must reg- discrepant financial results correlates with higher ister with the SAIC at the time of establishment odds of receiving a grant. and must submit annual inspection documents to maintain their legal status (Company Law, 2006; SAIC, 2006). These documents include detailed DATA AND SUMMARY STATISTICS financial statements that must be approved by registered accounting firms. Financial data Several features of the Chinese accounting system make it appropriate to compare the MOST and Our data include 467 firms applying for an Innova- SAIC financial statements. First, China has adopted tion Fund (Innofund) grant from five cities in two a “unified accounting system” statute, with strict Chinese provinces. Innofund is modeled after the guidelines regarding how firms must prepare and U.S. Small Business Innovation Research (SBIR) file their financial documents (Accounting Law, program: It was specifically created to overcome a 2000). This unified accounting system is overseen “market failure” by supporting innovative and com- by a single national agency, the Ministry of Finance mercially appealing projects at an early stage of of the State Council, and the country’s accounting development.3 The Ministry of Science and Tech- laws mandate its implementation. Second, the fiscal nology (MOST) administers Innofund. Grantees year for all companies in China is statutory: Fiscal are awarded a sum of 500,000–1 million yuan years must correspond to the solar year of January RMB from the MOST, with a guaranteed match of 1 to December 31. Third, Chinese Accounting Law 50–100 percent from local government.4 For young explicitly prohibits firms from creating different Chinese companies, a grant of this magnitude, with books or changing accounting measures in reports no dilution to equity, is substantial. We chose to prepared for different users. For instance, the fol- study the Innofund because all applicants must sub- lowing acts are explicitly singled out as violations mit detailed financial statements to apply for grants of accounting law, “ … the measures for accounting and for the same set of companies at the same points arrangement are arbitrarily changed; … the basis for in time, we were able to retrieve a second set of preparing financial and accounting reports provided financial statements submitted to a different state to different users of accounting documents is incon- agency. sistent” (Article 42). The Accounting Law further Chinese law unambiguously states that all finan- states, “Except for the statutory account books, a cial statements must be compiled according to the company shall not set up other account books” same accounting rules. Because there is no legiti- (Article 172).5 To facilitate the implementation of mate reason for discrepancies between the two sets of books and because there are very clear incen- 5The following sections of the law also are relevant. Article tives to directionally (see below) report different 16 requires “All economic and business transactions that take results to the two agencies, we can use the pres- place in a unit shall be recorded and calculated in the account ence of discrepancies to measure fraud. Another books set up according to law, and no unit may, in violation of the provisions of this Law and the State’s unified accounting strength of the Innofund dataset is its coverage of system, set up privately any other account book for recording and calculating such transactions.” Article 20 states, “Financial accounting statements shall be prepared on the basis of the 3 examined and verified records of the account books and the related See http://www.innofund.gov.cn/english/02_fund_nature.htm materials information and comply with the requirements set by 4The exchange rate on June 10, 2014, was $1 = 6.22 yuan RMB. this Law and the State’s unified accounting system.” Article 25

Copyright © 2015 John Wiley & Sons, Ltd. Strat. Mgmt. J., 37: 2658–2676 (2016) DOI: 10.1002/smj 2664 T. Stuart and Y. Wang the Accounting Law, the State Council also pub- a measure of profit, we use earnings before taxes. lishes an accounting principles guideline with more The Y-axis in Figure 1 represents the profit number than 150 pages of instruction for preparing and reported to the MOST while the X-axis portrays recording accounting statements. the profit number report to the SAIC for the same Even though the MOST and SAIC books are sub- company. Firms that report identical numbers to mitted by the same firms covering the same account- the two authorities populate the 45-degree line. ing period, and are required by statute to report Observations above the line indicate that greater the same information, there are many instances profits are reported to the MOST than to the SAIC. of significant discrepancies in the two sets of Figure 1 illustrates a high level of inconsistency. books. Moreover, most firms have clear incentives Specifically, 60 percent of the firms in our data to underreport their financial performance to the reported different profit numbers to the MOST and local SAIC for the purpose of tax avoidance. While SAIC. The incidence of discrepancy falls by just not directly responsible for tax collection, the local five percent when we allow for “rounding error” SAIC collaborates with the State Administration as high as 20 percent of reported profits. Thus, of Taxation (SAT) to conduct joint inspections of 55 percent of firms report a profit discrepancy and coordinate their administrative actions against exceeding 20 percent between the two books. tax evaders (Tian, 2012). Given the close relation- To measure the discrepancy in total reported ship between these agencies, there is an incentive to profit between the MOST and SAIC books, we underreport profits to the SAIC. subtract the latter from the former and label the In contrast to the SAIC’s problem of understate- difference, Profit Gap. Equating a profit gap to ment of profits, the MOST is likely to receive over- fraud, Panel A in Table 1 illustrates the percentage statements of true profits. While modeled after U.S. of fraudulent cases using different levels of dis- SBIR grants to promote innovation, the Innofund crepancy as cut points. (For the remainder of the expressly considers financial performance in its article, we refer to a company as “fraudulent” when evaluation of candidates. Each applicant is rated by its profit numbers are discrepant between the two a panel of financial and technical experts. Regard- sets of books.) less of other dimensions of merit, firms that receive a low financial rating are eliminated from further The inconsistencies between the MOST and consideration. Thus, even if an applicant’s techni- SAIC books cannot be explained by “creative cal merit is deemed to be outstanding, it is dismissed accounting.” Panel B in Table 1 shows that the from the competition if it fails to meet certain finan- mean value of profit reported in the MOST books cial thresholds. is 1.392 million yuan RMB, which is a remarkable The average grant-winning firm receives 3.12 times greater than the mean value reported to 1.18 million yuan RMB from the MOST and local the SAIC, 0.338 million yuan. On average, firms in governments.6 This is a sufficiently large infusion our sample report 1.054 million yuan in additional for the capital-constrained, early-stage technology profits in their filing to the MOST, relative to their companies in our sample that the applicants have filing to the SAIS. Firms that switch from claiming an incentive to take steps to secure a grant. to be loss-making to profit-making also are reveal- For the firms in our sample, Figure 1 illustrates ing. Specifically, 43.47 percent of firms stated that the distribution of total profit across the MOST they made zero or negative profits in their SAIC and SAIC books. We focus on total profit because filing, but only 18.42 percent reported nonpositive interviewees informed us that the Innofund specif- profits in their MOST filing. ically evaluates profitability metrics and favors Without knowledge of the actual profit level of a profitable companies in its funding decisions. As firm (i.e., we have no way to know whether either reported profit figure is accurate), we cannot calculate the relative magnitude of misrepresentation by further requires firms to “confirm, calculate and record assets, debts, owners’ equities, revenues, expenses, costs, and profits weighing the discrepancy between the two sets of in accordance with the provisions of the uniform accounting books as a percentage of true profit. Instead, in the system of the State on the basis of the economic transactions and multivariate regressions, we control for firms’ reg- operational matters which actually occur.” istered capital base and employee headcount. For 6In 2011, Innofund supported 6,545 projects with 3.77 billion yuan RMB in funding, and these funds were augmented with an reasons we describe below, these two covariates are additional 4 billion yuan RMB provided by local governments. the most reliable indicators of firms’ true sizes.

Figure 1. Total profit reported in two books. This figure plots, for financial statements that are matched tothesame firm at the same time, the total profit (earnings before taxes) the firm discloses to the Ministry of Science andTechnology (MOST) and the profit it discloses to the State Administration of Industry and Commerce (SAIC). The Y-axis reports the MOST profit number, and the X-axis reports the SAIC profit number for the same firm. Firms that disclosethesame number to the two authorities fall on the 45-degree line. Observations above the line indicate that the corresponding firm reports a higher profit to the MOST than it does to the SAIC.N = 467.

Table 1. Magnitude and prevalence of profit manipulation among Chinese SMEs

Panel A: Distribution of firms with different magnitudes of total profit discrepancy

(ProfitMOST – ProfitSAIC)/(ProfitMOST + 1) Percentage (%) ProfitMOST – ProfitSAIC Percentage (%) ≥0.05 58.67 ≥10 k yuan 59.74 ≥0.10 57.39 ≥50 k yuan 57.39 ≥0.20 54.82 ≥100 k yuan 56.53 ≥0.30 53.10 ≥500 k yuan 43.90

Panel B: Comparison of total profits across two accounting books (unit: million yuan) Mean Median Std. dev.

Total profit reported to the MOST 1.392 0.812 2.120 Total profit reported to the local SAIC 0.338 0.045 1.723 Total profit gap between the two books 1.054 0.344 1.790

This table describes the discrepancy in total profit, defined as earnings before tax, reported to the Ministry of Science and Technology (MOST) and the State Administration of Industry and Commerce (SAIC). Panel A is the proportion of firms that report profits to the MOST and SAIC that are discrepant by more than the percentages (raw numbers) given in the first (third) column. Panel B reports descriptive statistics in total profits reported in the two books. The exchange rate on June 10, 2014, was$1 = 6.22 yuan RMB. N = 467.

Political connections They provide detailed information on founders’ We have hand-coded the resumes of all company educational background, former employers, major founders to construct a measure of political ties. career achievements, and recognitions from the These resumes are submitted to the Innofund as government. We define an indicator variable, Polit- a mandatory part of the grant application process. ical Connection, which denotes that one or more

Table 2. The magnitude and occurrence of profit manipulation, by firm type

(1) Magnitude of profit (2) Existence of nontrivial gap, in million yuan profit gap, dummy Obs. Mean Std. error Mean Std. error

Panel A: Connected vs. unconnected firms Firms with no political connection 365 0.882 0.088 0.496 0.026 Firms with political connection 102 1.667 0.200 0.735 0.043 Difference between two groups −0.785*** 0.197 −0.239*** 0.054 Panel B: Organizational equity holder (OEH) vs. non-OEH firms Firms w/o organizational equity holder 287 1.095 0.103 0.599 0.029 Firms with organizational equity holder 180 0.988 0.140 0.467 0.037 Difference between two groups 0.106 0.170 0.133*** 0.047 Panel C: VC vs. non-VC invested firms Firms with no VC investor 407 1.027 0.087 0.550 0.025 Firms with VC investor 60 1.235 0.247 0.533 0.065 Difference between two groups −0.208 0.248 0.017 0.069

Asterisks denote significance levels of two-tailed test: +p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001. This table shows the occurrence and magnitude of fraud by firm characteristics. The magnitude of profit manipulation is defined to be the difference of total profit filed in the MOST and SAIC booksMOST (Profit – ProfitSAIC). A firm has a nontrivial profit gap when theMOST and SAIC profit numbers it reports are discrepant by more than 20 percent (i.e.,(ProfitMOST – ProfitSAIC)/(ProfitMOST + 1) > .20). Panel A compares politically connected to unconnected firms. Panel B compares firms with organizational equity holder with firms without organizational equity holder. Panel C compares firms with venture capital (VC) investors to those without them. company founders, (1) previously worked in the organization, such as another corporation or a uni- Chinese government, or (2) once held membership versity. Second, Venture Capital Investor is 1 if one in the People’s Congress (PC) or the Chinese Peo- or more of the organizational equity holders is listed ple’s Political Consultative Conference (CPPCC). as a VC firm by the State Development Planning By this definition, 102 firms (21.84% ofthe Council and has a profile on the website of China’s sample) qualify as politically connected. Using a premier VC/PE information provider, Zero2IPO. similar definition, other studies have found that Panels B and C of Table 2 report cross tabulations politically connected firms enjoy a set of advantages showing fraudulent reporting between firms, with over nonconnected firms (e.g., Ang and Jia, 2014; and without Organizational Equity Holders and VC Li and Zhang, 2007). Due to their privileged posi- investors. In neither case is there a statistically sig- tions in Chinese society, these politically connected nificant, bivariate association between the magni- entrepreneurs often are called, “red capitalists.” tude of the profit gap companies report and their Panel A of Table 2 examines variation in fraud- ownership structures. However, when we examine ulent reporting between politically connected the binary indicator of a reported profit discrepancy, and unconnected firms. On average, connected companies with Organizational Equity Holders are firms have an additional three quarters million in statistically less likely to report a profit discrepancy profit discrepancy relative to unconnected firms (46.7% vs. 59.9%). (1.667 million yuan vs. 0.882 million yuan, respectively). Panel A also demonstrates that a far higher percentage of connected firms report discrepant Other entrepreneur and firm characteristics profits (73.5% vs. 49.6%). Both bivariate differ- We coded demographic and educational informa- ences are statistically significant at the p < 0.001 tion for all entrepreneurs in the sample. Founder level. Age is the age of a firm’s key founder in the year of its Innofund application. Founder’s Education is a four-category measurement of the key founder’s Ownership structure highest level of education. We create two dummy variables describing owner- State Incubator and Hi-Tech Zone are two ship structure. First, Organizational Equity Holder dummy variables indicating whether a firm is is 1 when any equity owner in the company is an located in a state-designated accelerator or a

Table 3. Profit manipulation and recipient of a MOST grant

(1) Recipient of the grant (2) Size of the grant Obs Mean S.E. Mean S.E.

Firms without nontrivial profit gap 211 0.469 0.033 28.152 2.172 Firms with nontrivial profit gap 256 0.598 0.031 39.609 2.171 Whole sample 467 0.540 0.230 34.433 1.563 Difference between two groups −0.128*** 0.046 −11.457*** 3.100

Asterisks denote significance levels of two-tailed test: +p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001. This table compares the recipient and size of the MOST−administered Innofund grant between firms with and without discrepant profit reports. A firm is defined as having a nontrivial profit gap when the MOST and SAIC profit numbers it reports differ bymorethan 20 percent. The unit for the size of the grant is 10K yuan RMB. high-technology zone. Firm Age is the number operate in the seven industries that are targeted by of years since establishment at the time of the the Innofund. These include IT and electronics, Innofund application. The firms in the sample were biotech and medicine, advanced materials, automa- relatively young, with an average age of five years. tion, new resources and environmental protection, We use the natural logarithm of a firm’s registered conservation and renewable energy, and high-tech capital (Registered Capitalln) and number of services. All other firms are treated as a residual employees (Employeesln) as proxies for size. Both category. variables are less likely to be manipulated than assets or other measures of size. In particular, as part of the submission process, Innofund applicants Innofund grant are required to submit a photocopy of their business From the Innofund website, we collected informa- licenses, which lists their registered capital. tion on all grant recipients. Of the firms in our sam- We also collect data on firms’ grant size requests ple, 54 percent received Innofund funding, which and technical endowments. The Innofund expressly is higher than the Innofund’s 45 percent acceptance favors applicants with at least some track record rate in the period of our data, 2005–2010. The of discovery. Because winning a grant is the major higher award rate in our sample reflects several motivation for firms to exaggerate their financial sampling choices. First, we excluded firms that performance to the MOST, those applying for larger were less than one year of age at the time of grants may report a greater profit discrepancy. In their Innofund application because these companies contrast, firms with superior technical endowments often have incomplete financial records. Second, may have a weaker incentive to exaggerate their we were only able to acquire the SAIC data for financials because they can rely on their merits to major metropolitan areas. The grant rejection rate win a grant. was higher among very young firms and those from To measure technical endowments, we construct nonmetropolitan locales. a Patents Applied variable. We specifically avoid The first panel in Table 3 shows that firms that using data on R&D spending as applicant firms have cook their books were 12.8 percent more likely to an incentive to misrepresent this quantity in their receive an Innofund grant than nonfraudulent firms. 7 applications. By contrast, the patent count is ver- This difference is statistically significant at the ifiable to evaluators because the MOST requires p < 0.001 level. Considering that 54 percent of firms actual copies of patents to be included in applica- in the overall sample received Innofund grants, the tions. Therefore, data on intellectual property filings advantage that profit-manipulating firms garner is should be accurate in our sample. meaningful. The second panel in Table 3 shows Finally, we control for industry and geographic that firms with discrepant financials on average location. Out of the 467 firms in the sample, 408 received 114.57K more yuan RMB in Innofund grants. Given that the median total profit reported by our sampled firms to the MOST (SAIC) is 7In fact, to qualify for consideration for an Innofund grant, firms are required to allocate a minimum of five percent of revenues to 812K (45K) yuan RMB, an additional 114.57K is a research activities. significant sum.

MULTIVARIATE RESULTS of a company’s patented inventions. As anticipated, companies with a strong patent portfolio generally We conduct the empirical analysis in two steps. are less likely to commit financial fraud. First, we examine the antecedents of fraudulent Looking at the columns in the table, we find a reporting. Next, we investigate the relationship strong, positive effect of political connections on between fraudulent reporting and the odds of receiv- fraudulent reporting. Column 1 in Table 4 indi- ing a MOST grant. cates that a connected firm reports an additional 590,000 yuan RMB in profit to the MOST relative to the SAIC. To contextualize this number, the median Political connections and financial data of the distribution of profits reported to the MOST manipulation was 812,000 yuan. In contrast, the median of the We begin by regressing firms’ profit discrepancy distribution of profits reported to the SAIC was only on political connections and control variables. 45,000 yuan. Thus, the estimated effect of political We report models with three specifications of the connections is almost three-fourths the size of the outcome variables, and we then present additional median profit reported to the MOST, and more than analyses that shed light on the differences in results 13 times the profit reported to the SAIC. across functional forms. The first panel in Table 4 Turning to the logistic regressions in Panel C, reports OLS regressions of the Profit Gap, the Column 9 shows a large effect of political connec- magnitude of the profit discrepancy reported to the tions on the likelihood that a company reports a MOST and the SAIC. To reduce the influence of nontrivial profit discrepancy between the two sets of extreme values, we Winsorized the Profit Gap at books. The parameter estimate indicates a striking, one percent. Panel B in Table 4 uses the natural 161 percent higher odds that politically connected logarithm of the Profit Gap to accommodate skew. firms commit fraud. As there are a small number of negative values in Profit Gap, the natural logarithm transformation drops six observations in the analyses.8 Last, we Organizational equity holders and financial run a logistic regression of Nontrivial Fraud, data manipulation which is defined to be a substantial profit dis- In Columns 2, 6, and 10 in Table 4, we add a crepancy between the MOST and SAIC books dummy variable: The firm has one or more outside ( ProfitMOST – ProfitSAIC > 0.20). ProfitMOST+1 organizational equity holders (OEHs). As we antic- Across the columns of Table 4, three control ipated in Hypothesis 2, in all three sets of regres- variables are significant in the majority of speci- sions, we find a negative correlation between profit fications. First, companies with higher headcounts manipulation and having an OEH. The effect is are more likely to cheat. Because the dependent strongly significant in the logit regressions of non- variable is not scaled by firm size, however, it trivial discrepancy (Panel C, Col 10), though it is is unsurprising that larger firms report more dis- shy of statistical significance in the OLS models crepant numbers. Second, firms applying for a (Panel A, Col 2, and Panel B, Col 6). The mag- larger grant are more likely to cook their books. nitudes suggest that companies with an OEH have Third, the payoff to fraud appears to be a function 42 percent lower odds of reporting a profit discrep- of merit. The potential gains from presenting mis- ancy than otherwise comparable firms. This accords leading information in financial disclosures depend with Hypothesis 2: Because they encourage trans- on the counterfactual response that an applicant parency or because they provide access to capital, expects if it discloses accurate information. In and therefore, weaken the incentive for companies general, the greater a company’s odds of winning a to cheat to acquire resources, the presence of organi- grant on the merits, the lower will be its incentive zational owners on a company’s capitalization table to cheat. The cleanest measure of merit for an discourages fraud.9 Innofund grant that is available to us is the number

9Given the positive association between politically connected 8We reran the analyses implementing a cubic root transformation entrepreneurs and fraud, it is natural to ask whether companies of the Profit Gap. This reduces skew, but does not require an with state-affiliated organizational equity holders behave differ- adjustment to 0-valued observations. Results are highly similar ently from other firms. We identified 81 firms with OEHs that are across these and logarithmic transformations. either state-owned enterprises or state-affiliated VCs (i.e., the VC

Copyright © 2015 John Wiley & Sons, Ltd. Strat. Mgmt. J., 37: 2658–2676 (2016) DOI: 10.1002/smj Who Cooks the Books in China, and Does It Pay? 2669 0.20. > 1.243 0.107 0.011 0.238 Full ) model 0.237** 0.793** − − − − )]). Models − − MOST SAIC Profit + – Profit 1.246 0.106 0.174 0.238** 0.713** )/(1 − − − − MOST SAIC (Profit – Profit + 1.419 0.057 0.002 0.000 0.069 0.219* 0.539* − − − − − − MOST oks (e.g., ln[1 0.741 0.137 0.006 0.203 e statement it disclosed to the State Administration of 0.217* hat is (Profit − − − − Political − connection OEH VC re ordinary least square (OLS) regressions, and the dependent +− ts. Robust standard errors are presented below the coefficients. 0.122 0.103 0.185 0.113 0.107 0.009 0.003 Full 0.026* 0.645* 0.131 model − − − − − +− 0.116 0.005 0.027* 0.681* 0.117 − − − − 0.104 0.081 0.006 0.024* 0.714* − − − − − 0.119 0.008 0.023* 0.587* − − 0.001. Political − connection OEH VC < p +− +− 0.01, *** 0.354 0.002 Full 2.139* 0.052 0.346 model < − − − p +− +− 0.05, ** < 0.342 p 2.232* 0.056 0.316 (0.188) (0.188)(0.355) (0.343) (0.068) (0.068) (0.122) (0.119) (0.255) (0.450) (0.266) (0.445) − − − − 0.10, * < p + 0.310 0.048 0.221 2.316* (0.181) (0.065) (0.244) − − − − Panel A: Winsorized OLS Panel B: Logarithm OLS Panel C: Logit (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) 0.348 0.044 1.981* (0.304) (0.302) (0.316) (0.317) (0.104) (0.104) (0.107) (0.108) (0.402) (0.411) (0.414) (0.417) (0.032) (0.032) (0.030) (0.030) (0.012) (0.012) (0.012) (0.011) (0.087) (0.086) (0.085) (0.087) (0.898) (0.916) (0.914) (0.916) (0.287) (0.293) (0.293) (0.292) (1.079) (1.097) (1.120) (1.147) (0.073)(0.106) (0.077) (0.080) (0.110) (0.079) (0.114) (0.108) (0.032) (0.044) (0.033) (0.034) (0.045) (0.034) (0.045) (0.044) (0.112) (0.183) (0.115) (0.118) (0.186) (0.188) (0.117) (0.185) (0.027) (0.027) (0.027) (0.027) (0.014) (0.013) (0.013) (0.013) (0.037) (0.035) (0.035) (0.037) (0.350)(0.006) (0.349) (0.347) (0.006) (0.347) (0.006) (0.006) (0.130) (0.002) (0.128) (0.130) (0.002) (0.131) (0.002) (0.002) (0.439) (0.007) (0.412) (0.420) (0.007) (0.007) (0.446) (0.008) (0.211) (0.207) (0.076) (0.075) (0.316) (0.317) − − Political − connection OEH VC ) 0.225 0.211 0.215 0.233 0.241 0.226 0.230 0.249 80.151 79.723 84.441 89.065 2 (Chi 2 Hi-tech zone 0.120 0.213 0.151 0.097 0.004 0.038 0.017 State incubator Fixed EffectsD.F.R Yes Yes 29 Yes 29 Yes 30 Yes 31 Yes 29 Yes Yes 29 Yes 30 Yes 31 Yes 29 Yes 29 30 31 Ln employeesPatent count 0.384*** 0.416*** 0.444*** 0.411*** 0.158*** 0.170*** 0.180*** 0.168*** 0.125 0.185 0.247 0.184 Ln reg. capital 0.056 0.091 0.070 0.067 0.006 0.020 0.012 0.010 All models include unreported controls for founders’ age, gender, and level of education as well as city, industry, and application-year fixed-effec Appl. grant sizeConstant 0.013* 0.014* 0.013* 0.013* 0.005* 0.005** 0.005* 0.004* 0.018* 0.017* 0.017* 0.018* in Panel C are logit regressions, and the DV is a dummy variable that equals to 1 if nontrivial profit discrepancy exists between the MOST and SAIC books, t Non-VC OEH This table examines how politicalvariable connections (DV) is and the ownership structure WinsorizedIndustry are (at and associated 1%) Commerce (SAIC). gap with Models in fraudulent in profit manipulation Panel between of B are profit alsothe financial data. OLS statement Models regressions, and the in DV a firmPanel is A a the reports natural logarithm to ofthe the Ministry profit gap of Science between and Technology the MOST (MOST) and th and SAIC bo Observations 467 467 467 467 461 461 461 461 467 467 467 467 Org. Equ. Holder (OEH) Table 4. Determinants of fraud Pol. connection 0.590** 0.602** 0.227** 0.233**Asterisks denote significancelevels 0.960** two-tailed of test: 1.009** VC OEHFirm age 0.001 0.005 0.006 0.119 0.071 0.042 0.024 0.189 0.073

Table 5. Determinants of fraud, extension

Panel A: Winsorized OLS Panel B: Logarithm OLS Panel C: Logit (1) (2) (3) (4) (5) (6)

Pol. connection 0.224 0.261 0.058 0.071 0.699 0.751 (0.392) (0.381) (0.143) (0.140) (0.538) (0.519) VC OEH 0.760+ 0.264+ 1.103* (0.446) (0.148) (0.521) VC-only OEH 1.011+ 0.354* 1.481* (0.532) (0.177) (0.723) VC & non-VC OEH 0.540 0.184 0.800 (0.520) (0.180) (0.609) Firm age 0.033 0.030 0.015 0.014 0.008 0.005 (0.046) (0.045) (0.018) (0.018) (0.057) (0.058) Ln reg. capital 0.018 0.037 0.019 0.026 −0.011 0.023 (0.147) (0.150) (0.057) (0.059) (0.192) (0.198) Ln employees 0.437* 0.455* 0.149* 0.155* 0.116 0.143 (0.183) (0.184) (0.068) (0.068) (0.280) (0.279) Patent count −0.031 −0.037 −0.016 −0.018 −0.167 −0.165 (0.047) (0.047) (0.016) (0.016) (0.138) (0.127) State incubator −0.462 −0.457 −0.184 −0.180 0.292 0.267 (0.613) (0.614) (0.192) (0.191) (0.607) (0.595) Hi-tech zone 0.304 0.332 0.059 0.068 0.207 0.227 (0.526) (0.524) (0.179) (0.179) (0.616) (0.609) Appl. grant size −0.009 −0.009 −0.005 −0.005 −0.005 −0.004 (0.010) (0.010) (0.004) (0.004) (0.017) (0.017) constant −2.167 −2.222 −0.693 −0.715 −3.594+−3.723+ (1.680) (1.671) (0.580) (0.578) (2.074) (2.086) Fixed effects Yes Yes Yes Yes Yes Yes D.F. 30 31 30 31 30 31 R2 (Chi2) 0.216 0.220 0.252 0.256 44.776 44.763 Observations 180 180 175 175 180 180

Asterisks denote significance levels of two-tailed test: +p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001. This table examines whether different ownership structures are associated with fraudulent manipulation of profit data. All observations have organizational equity holders (OEH), and the omitted group is firms with only non-VC OEHs. All models include unreported controls at the firm and entrepreneur levels. Robust standard errors are presented below the coefficients.

Venture capital investors and financial data discrepancy and on the probability of reporting a manipulation profit gap is driven entirely by non-VC investors. Moreover, in contrast to Hypothesis 3, we find In Columns, 3, 7, and 11, we separate owners by type. The omitted category is companies owned tentative evidence that VC-backed companies in the entirely by individual shareholders, which we sample actually are more likely to commit fraud. compare to (1) companies with organizational To illustrate this, Table 5 shows regressions from equity holders, none of which are VCs (Non-VC the subsample of companies with one or more orga- OEH); and (2) companies with VC investors (VC nizational equity holders. This table demonstrates OEH). Surprisingly, we find that the negative effect that compared to companies that have non-VC of having OEHs on both the magnitude of the profit organizational investors, VC-backed firms are more likely to report discrepant profits and also to report a larger profit gap. Therefore, the presence of a ven- firms in which the state is either a limited partner or a general part- ture investor correlates with an increase in a portfo- ner in the fund). T-tests show no significant difference between firms with state-affiliated organizational owners and the remain- lio company’s proclivity to cook its books. ing firms in the sample. There is, however, weak evidence that, Why might VC-backed companies be more prone among only firms with organizational equity holders, those with to fraud? Although VCs have been shown to pro- state-affiliated investors are more likely to manipulate their financial data. However, this difference is statistically significant only mote good corporate governance in the United at the 0.10 level. States, their role in this regard is more ambiguous

Copyright © 2015 John Wiley & Sons, Ltd. Strat. Mgmt. J., 37: 2658–2676 (2016) DOI: 10.1002/smj Who Cooks the Books in China, and Does It Pay? 2671 in other countries (Cumming and Johan, 2013; Does fraud pay? Cumming and Walz, 2010). Bruton and Ahlstrom Table 6 examines the correlation between financial (2013) compares the conducts of Chinese VCs to fraud and receiving an innovation grant from the U.S. counterparts. These authors find that ven- MOST. Panel A reports logit regressions of selec- ture investors in China employ different selection tion for a grant. Panel B reports OLS regressions strategies to identify investments and different con- of the size of grant that the MOST allocates to tract terms to manage them. Likewise, White et al. each applicant firm. We estimate these regressions (2005) find that native VC firms in China areless to determine whether fraud pays. likely than U.S. firms to be activist investors. Sim- Column 1 shows that firms with political connec- ilar patterns have been observed in other emerg- tions and those with VC investors are much more ing economies (Bruton, Ahlstrom, and Singh, 2002; likely to receive Innofund grants. Though there Cumming, Schmidt, and Walz, 2010). may be omitted variables that affect VC funding With specific regard to fraudulent acts, VCs decisions and Innofund selection, this effect is may be well positioned to engage in rent-seeking particularly large: The coefficient on VC invest- behaviors in emerging economies. First, VCs enjoy ment indicates that a company backed by VCs has a halo effect with government officials because of a approximately 162 percent higher odds of being widespread belief that they have been central agents awarded an Innofund grant than does an otherwise in the success of the U.S. technology sector. In similar firm with no VC on its capitalization table. consequence, governments often adopt pro-VC pol- Technological merit also appears to be a relevant icy concessions, including expedited and reduced criterion in Innofund’s selection strategy. The coef- exposure to the regulatory process.10 Second, many ficient for firms’ patent count varies between 0.128 VC firms have close ties to the state through their and 0.152. This implies that each additional patent limited partners, which can include pension funds is associated with approximately 13.7–16.4 percent and state agencies (White et al., 2005). Third, higher odds of being awarded a MOST grant. rather than investing in one or two firms, VCs have Columns 2–4 examine whether reporting dis- a portfolio of companies under management. As a crepant results to the two agencies correlates with group, VCs and their portfolio companies can har- the probability of being selected for a grant. Across vest the benefits of economies of scale in building the three columns of results, we find a positive asso- relationships with the state. Therefore, a plausible ciation between financial data manipulation and explanation for the positive effect of the presence receipt of governmental grants, though the results of a VC investor on fraud is that VC-backed firms exhibit a notable nuance, which we describe below. are, like the politically connected companies in the Column 2 includes the magnitude of the profit sample, better shielded from judicial scrutiny.11 discrepancy between the MOST and SAIC books as an explanatory variable. Controlling for various observable firm and entrepreneur-level character- 10In 1999, China’s State Council dispatched a circular to encourage state ministries and local governments to take steps to develop istics, we find no effect of the continuous measure China’s VC industry for the creation of an innovation-based econ- of profit discrepancy. In Column 3, however, we omy. This was a high-profile policy guideline and was jointly include a dummy variable that is set to 1 if a drafted by the Ministry of Science and Technology, the State Development Planning Commission, the State Economic and nontrivial profit gap exists between the MOST and Trade Commission, the Ministry of Finance, the Central Bank SAIC books. Using a dummy variable to indicate of China, the State Administration of Taxation, and the China the existence profit discrepancy, we find a positive, Securities Regulatory Commission. Enticed by the prospect of VC’s role in nurturing the growth of high-tech firms, many statistically significant association between cooking local governments have offered very generous inducements to the books and receiving a grant. When other covari- VC firms to invest locally. For example, one city in Anhui ates are set to their means, fraudulent firms have Province provides rent-free office spaces, tax reductions (for up to three years), state procurement of services and products, state approximately 66.7 percent higher odds (or a 10.8% protection against VC losses (up to 10 million yuan RMB), and higher chance) of receiving an Innofund grant.12 streamlined regulatory procedures. For more detailed information, see: http://www.whjhq.gov.cn/readnews.asp?id=25422. 11We have conducted extensive, supplemental analyses of the venture capital ownership effect. In general, we find no differences suggestive evidence that portfolio companies of larger venture in the effects when we distinguish among three types of venture firms are more likely to commit fraud. investors: those with limited partner or general partner ties to the 12We should caution that a causal interpretation of these findings is State, foreign VCs, and all other investors. However, we do find premature because it remains possible that an unobserved variable

Table 6. Financial data manipulation and receipt of governmental grants

Panel A: Logit models on receiving Panel B: OLS models on size an Innofund grant or not of the Innofund grant received (1) (2) (3) (4) (5) (6) (7) (8)

Size of profit gap 0.090 1.637+ (0.064) (0.968) Existence of nontrivial gap 0.511* 8.739* (0.229) (3.438) Top-quartile of reported profit gap −0.204 5.610 (0.242) (4.335) Fraudulent, non-top quartile 0.554* 10.966** (0.236) (3.818) Pol. connection 0.542* 0.495+ 0.454+ 0.520+ 10.845** 9.859* 9.272* 9.536* (0.270) (0.273) (0.276) (0.278) (3.829) (3.899) (3.886) (3.889) Non-VC OEH −0.022 0.003 0.052 0.045 −2.358 −1.792 −1.018 −0.700 (0.247) (0.248) (0.252) (0.249) (3.605) (3.619) (3.651) (3.633) VC OEH 0.962* 0.970* 0.973* 0.962* 10.691+ 10.575+ 10.603+ 10.285+ (0.413) (0.415) (0.413) (0.420) (5.759) (5.693) (5.649) (5.661) Firm age 0.023 0.024 0.025 0.020 0.350 0.354 0.373 0.342 (0.036) (0.035) (0.035) (0.035) (0.498) (0.493) (0.491) (0.490) Registered capital, logged 0.124 0.116 0.131 0.151 2.335 2.224 2.483 2.692+ (0.113) (0.113) (0.113) (0.114) (1.610) (1.612) (1.600) (1.611) Employee number, logged 0.077 0.047 0.064 0.050 1.015 0.341 0.710 0.367 (0.163) (0.166) (0.164) (0.164) (2.339) (2.390) (2.316) (2.318) Patent count 0.128+ 0.135+ 0.152+ 0.148+ 1.955* 2.041** 2.235** 2.256** (0.076) (0.077) (0.078) (0.079) (0.777) (0.780) (0.789) (0.799) State incubator 0.056 0.085 0.046 0.011 2.636 3.215 2.515 2.676 (0.424) (0.426) (0.434) (0.436) (5.901) (5.932) (5.970) (5.964) Hi-tech zone −0.397 −0.397 −0.358 −0.238 −7.184 −7.343 −6.788 −5.552 (0.391) (0.392) (0.396) (0.403) (5.161) (5.189) (5.291) (5.439) Applied grant size 0.005 0.004 0.004 0.004 0.295*** 0.274*** 0.271*** 0.268*** (0.006) (0.006) (0.006) (0.006) (0.078) (0.077) (0.079) (0.078) Constant 0.453 0.641 0.300 0.360 10.265 13.767 7.601 8.707 (1.143) (1.158) (1.162) (1.166) (16.404) (16.381) (16.446) (16.523) D.F. 31 32 32 33 31 32 32 33 Chi2 60.259 61.561 62.698 62.901 R2 0.188 0.193 0.201 0.203

Asterisks denote significance levels of two-tailed test: +p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001. N = 467. This table examines the association between firms’ manipulation of profit data and their receipt of aMOST−funded innovation grant. All regressions in Panel A are logits, and the dependent variable = 1 if a firm receives an Innofund grant. All regressions in Panel B are ordinary least square (OLS) regressions, and the dependent variable is the size of the Innofund award received (unit = 10K RMB). All regressions include unreported controls for founders’ age, gender, and level of education. There also are controls for firm location in accelerator or high-tech zone as well as city, application year, and industry sector fixed-effects. Robust standard errors are presented below the coefficients.

Given the inconsistent results between the con- we explore the functional form of the relationship tinuous Profit Gap (Col 2) and the dummy vari- between the reported profit gap and the probability able specification (Col 3), in Column 4 of Table 6, of winning an Innofund grant. Specifically, the Col- umn 4 regression includes a three-category spline: an indicator for the top quartile of the distribution influences both companies’ propensity to commit fraud andto of the MOST-SAIC profit gap,13 an indicator for a receive an innovation grant. We believe this is unlikely given the functional form described below. Though we cannot claim that the specifications are persuasive, we have run Model 3 in Table 6with selection-on-observables estimators, and we continue to find that 13( To be concrete,) we rank all the relative ratios fraud predicts Innofund grants. Results are available on request. ProfitMOST – ProfitSAIC of profit data manipulation and create an ProfitMOST +1

Copyright © 2015 John Wiley & Sons, Ltd. Strat. Mgmt. J., 37: 2658–2676 (2016) DOI: 10.1002/smj Who Cooks the Books in China, and Does It Pay? 2673 reported profit gap that is below the 75th percentile, coverage and a much-enhanced corporate reputa- and an omitted category comprising all companies tion in the eyes of potential customers, the state, that report consistent profit numbers to the MOST and would-be investors. and SAIC.14 Here, we find an interesting relationship: This specification sheds light on the insignificant DISCUSSION AND CONCLUSIONS coefficient on the linear Profit Gap in Column 2. Specifically, we believe that the financial evaluators Using a unique dataset with two theoretically iden- at the MOST often are able to detect fraud when it is tical sets of financial books filed by the same blatant, which occurs when the profit number is out set of companies in the same period of time, we of whack in relation to other information available examine the determinants of fraudulent financial about a focal company. In this and other regression reporting among an economically significant but (e.g., Model 8 in Panel B), we consistently find that as-yet understudied population of companies: pri- companies that report a profit number to the MOST vate, technology firms in China. These unique data that is highly misaligned with the number reported enable us to avoid the common selection bias in to the SAIC are no more likely to obtain a grant research on corporate fraud, which occurs when than are companies that report consistent profits researchers must rely on regulatory actions or media in the two sets of books. Therefore, the MOST coverage to identify fraudulent conduct. To the best selection committee appears to be able to identify of our knowledge, this study is the first to document many of the egregious instances of manipulation the prevalence of fraudulent reporting among the of financial data, and to eliminate such firms from high growth, private-company sector in China. We consideration for a grant. find strong evidence that politically connected firms Finally, Panel B in Table 6 examines the associ- and highly suggestive evidence that venture-backed ation between fraudulent reporting and the size of companies are prone to commit fraud. Conversely, the grant the MOST allocates to each applicant. We companies with non-VC, organizational (vs. indi- find a similar pattern among the control variables vidual) investors are less likely to report discrepant as reported in Panel A; namely, firms with political financial results. We also find that firms that report connections, VC investors, and patent holdings discrepant financial statements are more likely to receive larger grants from the MOST. In Columns 6 receive government innovation grants. and 7, we observe positive, statistically significant This study builds on prior literature on the ben- associations between the magnitude and existence efits of political connections in emerging markets. of profit manipulation and the size of Innofund In particular, we demonstrate a new means through grants. In Column 7, the coefficient for profit which government ties create an advantage. Past manipulation is 8.74, suggesting that switching studies have documented that state officials favor an otherwise similar firm from being honest to politically connected firms in resource allocation cheating increase its Innofund award size by 87.4k and judicial decisions. We take an additional step yuan RMB. However, the predicted financial gains here, showing that even when applicants are evalu- almost surely underestimates the full benefit of the ated on their merits, connected firms still are advan- award, which include significant public relations taged in gaining access to state-controlled resources because they are better positioned to manipulate their financial disclosures to (appear to) perform well in merit-based evaluations. This shines a new indicator variable that a company’s profit data manipulation ratio light on the advantages of political connections. is in the highest quartile of this distribution. 14This split of data is based on our interviews with the Innofund We also present novel results regarding the role officials and external experts. These interviews suggest that they of venture capital in developing countries. Venture are well aware that applicant firms have incentives to inflate their capital in the United States is known to enhance financial results to increase their chance of receiving agrant. This is why the Innofund requires each application file to be corporate governance practices at portfolio com- graded by a financial expert, who has the power to reject any panies. In China, we find suggestive evidence to applicant suspected of fraud. Our interviewees also suggested the contrary. Although we obviously cannot offer that the financial experts pay particular attentions to firms whose financial statements appear to be “particularly out of line.”We direct evidence to this effect, our empirical results speculate that financial experts are more likely to deem fraudulent are consistent with the possibility that VCs in firms that significantly exaggerate their financial results. China encourage profit-seeking behavior even when

Copyright © 2015 John Wiley & Sons, Ltd. Strat. Mgmt. J., 37: 2658–2676 (2016) DOI: 10.1002/smj 2674 T. Stuart and Y. Wang it crosses the boundary of accounting laws. The even deleterious impact (e.g., Gorg and Strobl, caveat to this finding is that there are only 60 2007; Lerner, 2009; Wallsten, 2000). While never venture-backed companies in the dataset and the before explored, one potential factor contributing majority of the investors in these firms are local, to this policy failure may be that a nontrivial frac- Chinese VCs, not Western VCs with offices in tion of the private-sector beneficiaries of state aid China. receive resources under fraudulent representations This project does have one particularly signifi- of merit. If fraud is widespread among early-stage, cant limitation, which we reiterate in conclusion: innovation-focused companies, its ubiquity may We have no information on firms’ “true” financials, distort the allocation of public-sector resources to and therefore, we cannot know with complete cer- the entrepreneurial sector (e.g., Wang and Li, 2014). tainty which companies are honest. When identical This could be particularly true in the developing financial statements are filed with the MOSTand world where corruption is widespread and political SAIC, we have assumed this occurs because a firm insiders are positioned to capture state-dispensed reports accurate data to both agencies. However, it resources for private gains (e.g., Hellman, Jones, is possible that firms which report consistent data and Kaufmann, 2003). simply are submitting the identical set of fraudulent financials to both state agencies. If this is the case,it obviously (and distressingly) implies an even higher ACKNOWLEDGEMENTS base rate of fraud than what we report, though it also would likely introduce some bias in the regressions. We thank Witold Henisz, two anonymous referees, While we cannot definitively rule out this possibil- Yasheng Huang, Donald Lessard, Jizhen Li, Donald ity, we do know that firms that do misreport almost Palmer, Jordan Siegel, Markus Taussig, Jun Wang, always do so in a manner that is consistent with and seminar participants at Boston University, the incentives: They either overreport to the MOST, Chinese University of Hong Kong, INSEAD, MIT underreport to the SAIC, or do both. Moreover, we Sloan, Northeastern University, Rice University, know that if firms do take the risk of reporting fraud- Tsinghua University, and the University of Illinois ulent results, they have a strong incentive to produce at Urbana–Champaign for helpful comments and discrepant results for the two agencies. criticisms. We also thank Kathy Hu, Qing Jiang, Though we wished for a different message, the Wendy Liu, Junyuan Wang, and Kevin Yin for pro- findings of the article reveal one form of poten- viding valuable research assistance, and the Ewing tial, firm-level performance difference in emerg- Marion Kauffman Foundation for financial support. ing markets: The set of firm-level characteristics All errors are our own. that lead to heterogeneity in the propensity to perpetrate fraud. Though these dynamics regrettably evolve on an unlevel playing field, the financial ben- REFERENCES efits of garnering nondilutive investment capital and the halo of a prominent, government-granted award Accounting Law. 2000. The accounting law of the peo- early in the lifetime of nascent enterprises may ulti- ple’s republic of China. Available at: http://www.mof. mately engender longer-term, positional advantages gov.cn/zhuantihuigu/kjfzth20/bjzl/200805/t20080519_ 20769.html (accessed 5 March 2012). to companies that cook their books. Ang YY, Jia N. 2014. Perverse complementarity: political Finally, we consider a potential, broader impli- connections and the use of courts among private firms cation of our findings. We believe that knowing in China. Journal of Politics 76: 318–332. the incidence of fraudulent behavior among small Baron JN, Burton MD, Hannan MT. 1999. Engineering companies is a matter for public policy. Across bureaucracy: the genesis of formal policies, positions, and structures in high-technology firms. Journal of Law, the world, myriad policy initiatives aim to develop Economics, and Organization 15(1): 1–41. vibrant, local entrepreneurship sectors. Some stud- Bruton G, Ahlstrom D. 2013. An institutional view of ies suggest that these state-initiated efforts could China’s venture capital industry: explaining the differ- be a major driver for growing the competiveness ences between China and the West. Journal of Business of high-tech industries in the developing world Venturing 18(1): 233–259. Bruton G, Ahlstrom D, Singh K. 2002. The impact of insti- (e.g., Hout and Ghemawat, 2010). While these tutional environment on the venture capital industry in policies are well-intended, their effects have been Singapore. Venture Capital: An International Journal limited and there is evidence suggesting possibly of Entrepreneurial Finance 4(3): 197–218.

Chaney P, Faccio M, Parsley D. 2011. The quality of Hochberg YV. 2012. Venture capital and corporate gov- accounting information in politically connected firms. ernance in the newly public firm. Review of Finance Journal of Accounting and Economics 51: 58–76. 16(2): 429–480. Chen G, Firth M, Gao D, Rui O. 2006. Ownership struc- Hout T, Ghemawat P. 2010. China vs the world: whose ture, corporate governance, and fraud: evidence from technology is it? Harvard Business Review 88(12): China. Journal of Corporate Finance 12: 424–448. 95–106. Correia M. 2009. Political connections, SEC enforcement Innofund. 2010. The application guideline for the Innova- and accounting quality. London Business School Work- tion Fund for small –technology-based firms. Available ing paper, London, UK. at: http://www.innofund.gov.cn/innofile/se_02_t03.asp Company Law. 2006. The company law of the people’s (accessed 5 March 2012). republic of China. Available at: http://tfs.mofcom.gov. Johan S, Zhang M. 2014. Reporting bias in private cn/article/ba/bl/201101/20110107349089.shtml equity: reporting frequency, endowments, and gover- (accessed 5 March 2012). nance. TILEC Discussion paper No. 2014-016. Tilburg Cumming D, Hou WX, Lee E. 2011. Corporate fraud, CEO Law and Economics Center (TILEC), Tilburg, The turnover and state ownership in China. Unpublished Netherlands. working paper, York University: Toronto, Ontario, Kaplan S, Stromberg P. 2003. Financial contracting the- Canada. ory meets the real world: evidence from venture Cumming D, Johan S. 2013. Venture Capital and Private capital contracts. Review of Economic Studies 70: Equity Contracting: An International Perspective. Aca- 281–315. demic Press: Waltham, MA. Lerner J. 2009. Boulevard of Broken Dreams: Why Public Cumming D, Schmidt D, Walz U. 2010. Legality and Efforts to Boost Entrepreneurship and Venture Capital venture capital governance around the world. Journal Have Failed—and What To Do about It? Princeton of Business Venturing 25: 54–72. University Press: Princeton, NJ. Cumming D, Walz U. 2010. Private equity returns and Lester RH, Hillman A, Zardkoohi A, Cannella AA. 2008. disclosure around the world. Journal of International Former government officials as outside directors: the Business Studies 41: 727–754. role of human and social capital. Academy of Manage- Faccio M. 2006. Politically connected firms. American ment Journal 51(5): 999–1013. Li H, Atuahene-Gima K. 2001. Product innovation strat- Economic Review 96(1): 369–386. egy and the performance of new technology ventures Firth M, Rui O, Wu WF. 2012. The effects of political in China. Academy of Management Journal 44(6): connections and state ownership on corporate litigation 1123–1134. in China. Journal of Law and Economics 54: 573–607. Li H, Zhang Y. 2007. The role of managers’ political net- Fisman R, Wei S. 2004. Tax rates and tax evasion: evidence working and functional experience in new venture per- from missing imports in China. Journal of Political formance: evidence from China’s transition economy. Economy 112: 471–500. Strategic Management Journal 28: 791–804. Fisman R. 2001. Estimating the value of political connec- Li K, Lu L, Qian J, Zhu JL. 2014. Enforceability and tions. American Economic Review 91: 1095–1102. the effectiveness of laws and regulations. Available at: Fulmer S, Knill A. 2012. Political contributions and http://ssrn.com/abstract=2053804. the severity of government enforcement. AFA 2013 Liebman B, Milhaupt C. 2008. Reputational sanctions in San Diego Meetings Paper. Available at SSRN: China’s securities market. Columbia Law Review 108: http://ssrn.com/abstract=2024069. 929–983. Gorg H, Strobl E. 2007. The effect of R&D subsidies on Liu H. 2011. A thunder against academic plagiarism: Li private R&D. Economica 74: 215–234. Liansheng is stripped of the National Science and Tech- Greve HR, Palmer D, Pozner JE. 2010. Organizations nology Progress Award. Beijing Daily 02 December. gone wild: the causes, processes, and consequences of Ma DL, Parish WL. 2006. Tocquevillian moments: char- organizational misconduct. Academy of Management itable contributions by Chinese private entrepreneurs. Annals 4: 53–107. Social Forces 85(2): 943–964. Harrison JR, Carroll GR. 2006. Culture and Demography Malesky E, Taussig M. 2009. Where is credit due? Legal in Organizations. Princeton University Press: Prince- institutions, connections, and the efficiency of bank ton, NJ. lending in Vietnam. Journal of Law, Economics, and Hellman JS, Jones G, Kaufmann D. 2003. Seize the state, Organization 25: 535–578. seize the day: state capture and influence in transition McMillan J, Woodruff C. 2002. The central role of economies. Journal of Comparative Economics 31(4): entrepreneurs in transition economies. Journal of Eco- 751–773. nomic Perspectives 16(3): 153–170. Hellmann T, Puri M. 2002. Venture capital and the pro- Morsfield SG, Tan CE. 2006. Do venture capitalists influ- fessionalization of start-up firms: empirical evidence. ence the decision to manage earnings in initial public Journal of Finance 57(1): 169–197. offerings? Accounting Review 81(5): 1119–1150. Hillman AJ, Zardkoohi A, Bierman L. 1999. Corporate National Bureau of Statistics. 1998. China Statistics Year political strategies and firm performance: indications of Book. China Statistics Press: Beijing, China. firm-specific benefits from personal services in the U.S. National Bureau of Statistics. 2010. China Statistics Year government. Strategic Management Journal 20: 67–81. Book. China Statistics Press: Beijing, China.

Nee V, Opper S. 2010. Political capital in a market quality on firm taxation and equity ownership. economy. Social Forces 88(5): 2105–2132. China Economic Net. 8 February 2012. Available at: Peng S, Li C, Li S. 2011. The story behind the first-ever http://www.ce.cn/macro/more/201202/08/t20120208_ removal of National Science and Technology Advance- 23052891.shtml (accessed 2 June 2012). ment Award. Southern People Weekly, March 3 Uzun H, Szewczyk SH, Varma R. 2004. Board composition 2011. and corporate fraud. Financial Analysts Journal 60(3): Peng M, Luo Y. 2000. Managerial ties and firm per- 33–43. formance in an emerging economy: the nature of a Wademan A. 2012. Double Paradox: Rapid Growth and micro-macro link. Academy of Management Journal Rising Corruption in China. Cornell University Press: 43(3): 486–501. Ithaca, NY. Pistor K, Xu C. 2005. Governing stock markets in transi- Wallsten SJ. 2000. The effects of government-industry tion economies: lessons from China. American Law and R&D programs on private R&D: The case of the Small Economics Review 71: 184–209. Business Innovation Research program. RAND Journal SAIC. 2006. Corporate annual inspection methods. of Economics 31(1): 82–100. Available at: http://qyj.saic.gov.cn/djfg/gz/200602/ Wang Y, Li J. 2014. Fraud and innovation: is there a t20060224_59600.html (accessed 5 March 2012). cheater’s discount? Working paper, Boston University, Schuler DA, Rehbein K, Cramer RD. 2002. Pursuing Boston, MA. strategic advantage through political means: a mul- White S, Gao J, Zhang W. 2005. Financing new technology tivariate approach. Academy of Management Journal ventures in China: system antecedents and institution- 45(4): 659–672. alization. Research Policy 34(6): 894–913. Siegel J. 2007. Contingent political capital and interna- Xin KR, Pierce J. 1996. Guanxi: connections as substitutes tional alliance: evidence from South Korea. Adminis- for formal institutional support. Academy of Manage- trative Science Quarterly 52: 621–666. ment Journal 39: 1641–1658. Tan J, Litschert RJ. 1994. Environment-strategy relation- Yu F, Yu Y. 2011. Corporate lobbying and fraud detection. ship and its performance implications: an empirical Journal of Financial and Quantitative Analysis 46: study of the Chinese electronics industry. Strategic 1865–1891. Management Journal 15(1): 1–20. Tian, X. 2012. Regular information exchange between the SAT and the SAIC will improve information